MiniMax H3 Image-to-Video: First-and-Last-Frame Transition Capability Demo

This demo uses the MiniMax H3 image-to-video pipeline to showcase the "first frame + last frame" control method: two still images lock in the starting and ending points of the shot, and the model fills in the motion, lighting, and pacing in between, outputting a short film with synchronized audio. The work uses only fields that actually exist in the schema—prompt describes the on-screen action and camera movement, duration controls clip length, resolution selects the output specification, and prompt_expansion_mode determines the prompt expansion strategy; the two images, first and last frame, are bound to first_frame_image and last_frame_image respectively (the schema specifies that when the first frame is missing, the last frame must be provided). In terms of composition, the first frame is a static close-up of the product, and the last frame is a composition of the same product from another camera angle; the two share color tone and material so the transition stays consistent. The prompt explicitly specifies a "slow orbit + push-in" camera language, so that the intermediate frames present an explicable motion trajectory. It is suitable for multi-shot rough cuts for short videos and ads, animating product images or illustrations into short films, and scenarios that require controlled before-and-after comparison or shot transitions. Research materials show that H3 covers image-to-video and first-and-last-frame control, and processes text and image inputs simultaneously within a unified context; the resolution specifications differ across modes, and the actual API parameters need to be confirmed according to the platform.